Comments on 'Out of Balance: Comparing Public and Private Sector Compensation Over 20 Years'

Keith Bender and John Heywood recently released a study concluding that although the average state and local government employee is paid more than the average private-sector worker, once individual characteristics such as factors like education, age and unionization are taken into account, the public-sector worker is actually paid less than his or her private-sector counterpart. There are at least two types of problems with the study — one methodological and one statistical — that cast serious doubt upon this conclusion.

Stay Engaged

Receive our weekly emails!

The methodological problem is that the thought experiment it tests is not the correct one if we are concerned with whether we can obtain our current level of public services at lower cost. The statistical technique used, known as regression analysis, indicates that if you had two people with the same characteristics, one in the private sector and one in the public sector, then the one in the public sector would get paid less. That may be well and good, but it doesn't really tell you whether public-sector employees are less well paid than private sector ones. If public employees are older because the use of tenure and civil service rules gives them more job security because they're harder to fire, then the public sector work force will be more expensive and it is not at all clear that this is due to increased productivity. 

The issue is not whether public sector workers are paid more (the data show that they are), but rather why they are paid more. One of the implications of the methodology of the study is that when a Ph.D. in philosophy in the public sector (e.g., a philosophy professor) earns less than a certified public accountant in the private sector, it means that public-sector employee is less well paid, since the professor has more education than the accountant and yet is paid less. However, it could very well be that a philosophy professor is better paid relative to the value of his production than is the accountant.  

There is a particular problem with the unionization effect. In the regression equation, the unionization variable increases wages by 16 percent. If all private sector employees were non-union and all public-sector employees were union, then holding everything else constant, public sector employees would be paid 16 percent more than private sector employees due to the fact that unions are stronger in the public sector. For 2008, the study's data shows that 50 percent of local workers and 39 percent of state workers were unionized, while only 9 percent of private sector workers are unionized. This indicates that a good portion of the higher wages in the public sector is due to unionization.

These examples illustrate a couple of technical problems. First, the authors use what is known as dummy variables to control for public employment. This controls for a change in the intercept of the regression, but not for differences in slopes. For example, suppose that the effect of age on salary is greater for public-sector workers than for private-sector workers because of civil service rules or some other reason. The dummy for private-sector workers would not pick that up, since it only adjusts for changes in the intercept of the regression, and the coefficient on the dummy variable would underestimate the effect of being in the public sector on wages. If the authors ran separate regressions for the private-sector and public-sector samples, one would have a better analysis of the situation.

This leads to a second major technical problem, that of heteroskedasticity. If two explanatory variables are correlated, such as unionization and public sector, then the regression technique will not be able to distinguish between the two effects. The effect of being in the public sector may be picked up in the regression coefficient for unionization. Since age, education and unionization seem to be correlated with the public-sector variable, the key estimate for the conclusion of the authors, the coefficient on their dummy variable public-sector worker is likely to be statistically biased.

Given the methodological and statistical problems of the Bender and Heywood study, their results do not show that public-sector employees are underpaid relative to equivalent employees in the private sector. In fact, for the taxpayer, this is not the relevant issue. The relevant issue is whether changes in government policy can yield the same or greater amount of public services with lower cost and this requires a different and more in-depth study.




Dr. Gary Wolfram, an adjunct scholar with the Mackinac Center for Public Policy, is William E. Simon Professor of Economics and Public Policy at Hillsdale College.